During a US government testing exercise, the Anthropic Mythos AI model identified vulnerabilities in highly sensitive computer systems within hours, as the company worked with US intelligence agencies through Project Glasswing to test how advanced AI could secure critical software before attackers exploit it.
The concern centers on the Anthropic Mythos model, which, during a US government testing exercise, identified vulnerabilities in highly sensitive computer systems.
A US official told the Associated Press that Anthropic worked with US intelligence agencies through Project Glasswing, an initiative built to secure critical software from the “severe” fallout that Mythos could pose to public safety, national security, and the economy.
AI Moves Faster Than the Rules
The test did not mean that the Anthropic Mythos model exploited the systems within hours. But the discovery alone was enough to raise a deeper question for governments, if AI can find hidden cracks this quickly, who decides where, when, and by whom it can be used?
On June 11, Democratic Senator Mark Warner placed the concern in public view during a hearing before the Senate Committee on Banking, Housing, and Urban Affairs.
“This tool broke into almost all of our classified systems, not in weeks but in hours,” he said, citing the head of the NSA and US Cyber Command, General Joshua Rudd.
That line explains why Washington is nervous. AI is no longer just a chatbot race between US tech giants. It is becoming part of national defense, cyber offense, critical software testing, and intelligence work. The same model that can help security teams patch old codes may also help attackers move faster.
That fear became clearer when the Five Eyes intelligence alliance warned that AI models able to launch major cyberattacks are months, not years, away. The group, made up of the US, UK, Canada, Australia, and New Zealand, urged governments and companies to “act now” before defenses fall behind.
“Frontier Al models are anticipated to exceed current industry expectations, fundamentally transforming both offensive and defensive cyber capabilities. The timeline is not years, it is months,” the alliance said.
This is the balance every government now faces. Strong AI can detect vulnerabilities earlier, improve critical software quality, watch unusual behavior, and respond faster to incidents.
But it can also lower the barrier for malicious actors, making fast attacks, more complex, and harder to stop.
Controls Become the New Battlefield
The US administration’s response has already created tension. Earlier this month, it ordered Anthropic to stop foreign nationals from using its latest models, Fable 5 and Mythos 5. Anthropic then disabled the models for all customers to comply with the directive, while saying it did not believe the steps were warranted by the security issue raised.
Anthropic Mythos model said the government believed it had found a way of bypassing, or “jailbreaking,” Fable 5. The company said the method found only a small number of previously known, minor vulnerabilities, and that other public models could also find them without a bypass.
Still, the concern is bigger than one company. More than 100 cybersecurity experts and leaders from companies including Adobe and Nvidia urged the government to lift the directive. They warned that removing strong cyber defense tools “without a good reason” could help US adversaries more than it hurts them.
Locking down models may reduce misuse, but it may also weaken defenders who need the best cyber defense tools to find flaws before attackers do. Leaving models open may support testing and innovation, but it may also place dangerous capabilities in the wrong hands.
Anthropic’s global compute push adds another layer, from the Anthropic Mythos model. The company is hiring data center teams in Australia and Japan as it expands AI capacity in the Asia-Pacific region.
It has said growth is straining its infrastructure, while also choosing democratic countries with secure legal and regulatory systems for large-scale investments.
This matters because AI power is not only about software. It’s about data centers, energy, chips, cables, skilled workers and trusted locations. Australia’s position in Five Eyes makes it attractive for sensitive compute, while Japan has stability, strong internet infrastructure and government interest in domestic AI.
More powerful models need more power, more access, and more oversight. If these systems become national security assets, then data centers themselves become part of the security map.
The answer is not to stop advanced AI power.
It’s to control it with serious testing, clear access rules, transparent risk checks, and human oversight. If AI can find flaws in hours, humans must make sure it is used to close doors, not open them.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Intelligent Tech sections to stay informed and up-to-date with our daily articles.
